Add dual-mode git-ai operation via global hooks and wrapper by svarlamov · Pull Request #530 · git-ai-project/git-ai

svarlamov · 2026-02-16T05:30:23Z

Summary

add hook-name binary dispatch so git-ai runs directly when invoked as core Git hooks
add managed global hooks install/uninstall with symlinked hook binaries under ~/.git-ai/git-hooks
add persisted forwarding state for previous global and repo-level hooksPath values
add repo-level hook self-heal from checkpoint flow and strict repo-over-global forwarding precedence
add internal git hook suppression guards to prevent recursive hook execution
add wrapper+hooks coexistence guardrails to avoid double-running managed logic
extend test harness with wrapper/hooks/both modes and isolated hook HOME/config
add hook-mode regression tests and precedence tests

Validation

RUSTFLAGS='-D warnings' cargo check --features test-support
cargo test --features test-support --test hook_modes
GIT_AI_TEST_GIT_MODE=hooks cargo test --features test-support --test reset
GIT_AI_TEST_GIT_MODE=hooks cargo test --features test-support --test stash_attribution
GIT_AI_TEST_GIT_MODE=hooks cargo test --features test-support --test cherry_pick
GIT_AI_TEST_GIT_MODE=hooks cargo test --features test-support --test rebase
GIT_AI_TEST_GIT_MODE=hooks cargo test --features test-support --test checkout_switch
GIT_AI_TEST_GIT_MODE=hooks cargo test --features test-support --test pull_rebase_ff
GIT_AI_TEST_GIT_MODE=hooks cargo test --features test-support --test merge_rebase
GIT_AI_TEST_GIT_MODE=hooks cargo test --features test-support --test subdirs
GIT_AI_TEST_GIT_MODE=both cargo test --features test-support --test reset
GIT_AI_TEST_GIT_MODE=both cargo test --features test-support --test stash_attribution
GIT_AI_TEST_GIT_MODE=both cargo test --features test-support --test cherry_pick
GIT_AI_TEST_GIT_MODE=both cargo test --features test-support --test rebase

Add hook-name binary dispatch and managed global hooks installation with persisted forward state for previous global/local hooksPath. Add repo-level hook self-heal from checkpoint flow, recursion suppression for internal git subprocesses, and wrapper+hooks coexistence safeguards to avoid double execution. Extend test harness with wrapper/hooks/both modes and add hook mode regression coverage.

git-ai-cloud · 2026-02-16T05:30:31Z

Stats powered by Git AI

🧠 you    ███████████░░░░░░░░░  53%
🤖 ai     ░░░░░░░░░░░█████████  47%

More stats

1.0 lines generated for every 1 accepted
0 seconds waiting for AI
Top model: codex::gpt-5.3-codex (1035 accepted lines, 1039 generated lines)

AI code tracked with git-ai

git-ai-cloud-dev · 2026-02-16T05:30:31Z

Stats powered by Git AI

🧠 you    ███████████░░░░░░░░░  53%
🤖 ai     ░░░░░░░░░░░█████████  47%

More stats

0.4 lines generated for every 1 accepted
0 seconds waiting for AI
Top model: codex::gpt-5.3-codex (1753 accepted lines, 648 generated lines)

AI code tracked with git-ai

devin-ai-integration · 2026-02-16T19:01:35Z

src/commands/git_hook_handlers.rs

+    }
+
+    let local_config_path = repo.path().join("config");
+    let current_local_hooks =
+        read_hooks_path_from_config(&local_config_path, gix_config::Source::Local);
+
+    // If no repo-level hooksPath is configured, do nothing.
+    let Some(current_local_hooks) = current_local_hooks else {
+        return Ok(());
+    };
+
+    if current_local_hooks.trim().is_empty() {
+        return Ok(());
+    }
+
+    let repo_state = repo_state_path(repo);
+    let current_is_managed = is_managed_hooks_path_str(current_local_hooks.trim());
+
+    if !current_is_managed {
+        save_hook_state(
+            &repo_state,
+            &HooksPathState {
+                previous_hooks_path: current_local_hooks.trim().to_string(),
+            },


🚩 execute_forwarded_hook captures stdout/stderr which changes timing behavior

The forwarded hook process has its stdout/stderr piped (Stdio::piped()) at lines 557-558, and the output is replayed after the process completes (lines 574-575). This means the forwarded hook's output is buffered entirely in memory and written after completion, rather than streaming in real-time. For hooks that produce progress output or interactive prompts, this could cause user-visible behavior changes compared to pre-installation behavior where git ran user hooks directly with inherited stdio. This is a design trade-off for clean output separation.

Was this helpful? React with 👍 or 👎 to provide feedback.

devin-ai-integration · 2026-02-16T19:02:52Z

src/commands/git_hook_handlers.rs

+            if is_post_commit_amend(&repo) {
+                // For --amend, post-rewrite (amend) owns rewrite mapping.
+                return 0;
+            }


🚩 Hooks mode skips commit_post_command_hook for amend — potential attribution loss

In the post-commit managed hook handler, amend commits are detected and skipped (line 1657–1660), with a comment that post-rewrite (amend) owns the rewrite mapping. However, in the wrapper path (git_handlers.rs:399-408), commit_post_command_hook runs unconditionally for ALL commits including amends — this is the function that processes the working log and creates fresh authorship notes from the current working state.

The post-rewrite amend handler (lines 1697–1711) only remaps the OLD commit's authorship note to the new commit via handle_rewrite_log_event(commit_amend(...)). If a user checkpoints new AI code and then runs git commit --amend, the working log has the new attributions, but in hooks mode only the old note is remapped — the new working log data may not be incorporated.

The test hooks_mode_amend_uses_single_amend_rewrite_event validates event counts in the rewrite log but does not assert on authorship attribution correctness after amend. This may be an intentional trade-off for the initial hooks mode implementation, but it represents a behavioral divergence from wrapper mode that could lose AI attributions during amend.

Was this helpful? React with 👍 or 👎 to provide feedback.

devin-ai-integration · 2026-02-16T19:03:08Z

src/git/repository.rs

+thread_local! {
+    static INTERNAL_GIT_HOOKS_DISABLED_DEPTH: Cell<usize> = const { Cell::new(0) };
+}
+
+pub struct InternalGitHooksGuard;
+
+impl Drop for InternalGitHooksGuard {
+    fn drop(&mut self) {
+        INTERNAL_GIT_HOOKS_DISABLED_DEPTH.with(|depth| {
+            let current = depth.get();
+            if current > 0 {
+                depth.set(current - 1);
+            }
+        });
+    }
+}
+
+/// Disable managed git hooks for internal `git` subprocesses executed through `exec_git*`.
+/// Use this guard around higher-level operations that already execute hook logic explicitly.
+pub fn disable_internal_git_hooks() -> InternalGitHooksGuard {
+    INTERNAL_GIT_HOOKS_DISABLED_DEPTH.with(|depth| depth.set(depth.get() + 1));
+    InternalGitHooksGuard
+}
+
+fn should_disable_internal_git_hooks() -> bool {
+    INTERNAL_GIT_HOOKS_DISABLED_DEPTH.with(|depth| depth.get() > 0)
+}


🚩 disable_internal_git_hooks guard is thread-local — background threads spawned inside hooks won't inherit it

The INTERNAL_GIT_HOOKS_DISABLED_DEPTH counter at src/git/repository.rs:26 is thread_local!. The guard created in run_pre_command_hooks (src/commands/git_handlers.rs:322) and run_post_command_hooks (src/commands/git_handlers.rs:399) only suppresses hooks on the current thread. If any hook logic spawns a background thread that calls exec_git, that thread won't have the guard active and its git subprocesses will run with hooks enabled. Currently, the push hooks (push_hooks::push_pre_command_hook) spawn a background thread to push authorship notes. Those threads call push_authorship_notes which uses exec_git. Since the push happens via exec_git on a different thread, hooks would NOT be suppressed for those internal git calls. In practice this is likely fine because the push operations (git push) are not hook-sensitive commands, but it's a design subtlety worth documenting.

Was this helpful? React with 👍 or 👎 to provide feedback.

…ssions

ci: run test matrix across os and git test modes

d0d572f

This comment was marked as resolved.

Sign in to view

svarlamov added 3 commits February 16, 2026 00:53

Fix hook env leak and ensure git_og bypasses hooks

f3bb9ec

Harden hook mode rewrite handling for squash/cherry/pull/reset/stash

04c7406

Format hook handler updates

e6e6f51

This comment was marked as resolved.

Sign in to view

Fix Windows hooks-mode test harness binary path

2264703

This comment was marked as resolved.

Sign in to view

svarlamov added 4 commits February 16, 2026 03:49

Optimize hook invocation startup for no-op hooks

184a360

Fix clippy warning in hook forwarding fast path

8a9e7a3

Fix repo hook self-heal repository resolution

325895b

Optimize hooks-mode performance for rebase-heavy workflows

e354af9

This comment was marked as resolved.

Sign in to view

Stabilize rebase suppression test across CI platforms

5ab22d8

This comment was marked as resolved.

Sign in to view

Restore suppressed hooks at rebase completion

496f9b2

This comment was marked as resolved.

Sign in to view

svarlamov added 4 commits February 16, 2026 12:57

Simplify rebase suppression to shared global hooks dirs

0e9462e

Fix hooks-mode amend detection for root commits

4a610d3

Remove globalGitHooks config alias

041a429

Fix formatting in feature flag config parsing

2f1f6cb

devin-ai-integration bot reviewed Feb 16, 2026

View reviewed changes

This comment was marked as resolved.

Sign in to view

devin-ai-integration bot reviewed Feb 16, 2026

View reviewed changes

add multi-mode benchmark script

e173850

Address Devin feedback on hook forwarding, push path, and amend regre…

fd93239

…ssions

Conversation

svarlamov commented Feb 16, 2026 • edited by devin-ai-integration bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Validation

Uh oh!

git-ai-cloud bot commented Feb 16, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

git-ai-cloud-dev bot commented Feb 16, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

This comment was marked as resolved.

Uh oh!

This comment was marked as resolved.

Uh oh!

This comment was marked as resolved.

Uh oh!

This comment was marked as resolved.

Uh oh!

This comment was marked as resolved.

Uh oh!

This comment was marked as resolved.

Uh oh!

devin-ai-integration bot Feb 16, 2026

Choose a reason for hiding this comment

Uh oh!

This comment was marked as resolved.

Uh oh!

This comment was marked as resolved.

Uh oh!

devin-ai-integration bot Feb 16, 2026

Choose a reason for hiding this comment

Uh oh!

devin-ai-integration bot Feb 16, 2026

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

svarlamov commented Feb 16, 2026 •

edited by devin-ai-integration bot

Loading

git-ai-cloud bot commented Feb 16, 2026 •

edited

Loading

git-ai-cloud-dev bot commented Feb 16, 2026 •

edited

Loading